How it runs
Import, score, decide, measure
no scraping, no vendor key
1 · Import
CSV
Three reports the account already produces: Associates earnings, the Creator Connections campaign history, and a product sheet with price, monthly sales, rating and creator-video counts.
Headers are matched by alias. Columns the parser does not recognise are listed back to you instead of being dropped quietly.
2 · Score
score-v1.0.0
Sales up, commission per sale up, creator competition down, adjacency to what already earns, budget still open, sample likely. Each contribution is stored with the row and shown on demand.
A signal without input is dropped from the weighted mean and lowers coverage. It is never counted as zero.
3 · Decide
bucket
The filmable list is capped and de-duplicated by parent ASIN, so a variation family cannot fill the queue. One message per brand groups the sample requests.
Outreach is a state machine: queued, sent, sample, filmed, placed, earning, dead. Nothing is auto-sent.
4 · Measure
30/60/90
Commission that arrives later is joined back onto the prediction. Precision at the top of the list is the number that matters, not how many rows were produced.
A ranking that never gets measured is a list, not a system.